Output URL Bidding

Panagiotis Papadimitriou, Hector Garcia-Molina, (Stanford University)

Ali Dasdan, Santanu Kolay(Ebay Inc)

2

Search Engine Results Page (SERP)

Organic Results

Sponsored Ads

Query

Sponsored Search Ads

3

Keyword bidding

Advertiser Search Engines

the social networklord of the rings

the matrixlotr III

... ...

# keywords = ~ 10K

KEYWORDS

4

Example SERPs

en.wikipedia.org/wiki/The_Social_Network

www.imdb.com/title/tt1285016/

www.imdb.com/title/tt133093/

en.wikipedia.org/wiki/The_Matrix

en.wikipedia.org/wiki/The_Lord_of_the_rings

en.wikipedia.org/wiki/The_Lord_of_the_rings

www.imdb.com/title/tt167260/

www.imdb.com/title/tt120737/

the social network

the matrix

the lord of the rings

lotr iii

5

Output URL bidding

Advertiser Search Engines

imdb.com AND wikipedia.org

# URLs = 2

URLs

6

Topics

• Implementation

• Evaluation

7

Implementation challenge

SERP

8

Alternative implementation solutions

1. Serialization

O: Organic Search ComponentS: Sponsored Search Component

Latency

Simplicity

SERP

2. Parallelization

SERP

O: Organic Search Comp. (Or + Op)Or’: Small replica of OrS: Sponsored Search ComponentV: Ad validation

More resources

No latency

9

Topics

• Implementation

• Evaluation

10

Bid language model

• Output Expression– e.g., a := (u1 u2) u3 (h1 h2)– u: URL• e.g., en.wikipedia.org/wiki/The_Social_Network

– h: host• e.g., en.wikipedia.org

11

How to evaluate/study output bidding?

• Use existing keyword campaigns to generate realistic output expressions to study

The social networklord of the rings

the matrixlotr III

……

Output Expression Generator

imdb.com AND wikipedia.org

12

INPUT: set of keywords R (from a keyword campaign)

OUTPUT: expression a that “covers” R, i.e., q R, matches(a, results of q)

Generator input & output

Candidate expressionsOutput

Expression

a1 := u1 u2 u3

a2 := u1 u4

a3 := u5

13

1. CompactnessContain few URLs

2. Spill minimizationDo not match “irrelevant” queries

Which expression to select?

Candidate expressionsOutput

ExpressionSize|a| spill(a,R)

a1 := u1 u2 u3 3 {}

a2 := u1 u4 2 {q5}

a3 := u5 1 {q4,q5, q6}

14

Output expression generation problem statement

• Query Set Output Covermin. γ|a| + (1-γ)|spill(a, R)|subj. to matches(a, q), q R

γ : regularization parameter• NP-hard to solve

Reduction from Set Cover, Red-Blue Set Cover (see paper)

• Developed Greedy Algorithm (see paper)

15

Is spill always “bad”?

• Example:– q1: lord of the rings

– q2: the matrix

– q3: the social network

– q4: …

– q5: lotr

– q6: … Output

ExpressionSize|a| spill(a,R)

a1 := u1 u2 u3 3 {}

a2 := u1 u4 2 {q5}

a3 := u5 1 {q4,q5, q6}

→(lord of the rings)

16

Spill may be good!

• Cluster queries Q using the bipartite graph (see paper)

• Divide spill(a, R) into: – positive: relevant – negative: irrelevant

OutputExpression

Size|a|

spill(a,R)

+ -

a1 := u1 u2 u3 3 {} {}

a2 := u1 u4 2 {q5} {}

a3 := u5 1 {q5} {q4, q6}

17

Experimental evaluation goals

• Recall output expression looks like– a := (u1 u2) u3 (h1 h2)

• Evaluation questions– URLs, hosts or mixed?– Specific or generic? (# conjuncts)– Long or compact? (# disjuncts)

• Comparison criteria– Compactness vs spill tradeoff– Positive vs negative spill

18

Experimental setup

• Dataset (from Yahoo query logs)– 2,251 ads– 13M queries, 63M URLs (7M hosts)

The social networklord of the rings

the matrixlotr III

……

Output Expression Generator

imdb.com AND wikipedia.org

γ - URLs- hosts- mixed

# conjuncts

19

Compactness vs spill tradeoff

• URLs, hosts or mixed?Mixed expression curves

dominate others

• Specific vs generic?2 conjuncts suffice

• Long or compact?Next slide…

20

Positive vs negative spill

• For |a|>60, more than 50% of spill is positive

• URLs, hosts or mixed?Mixed expression curves

dominate others

• Long or compact?|a|=70-80 suffice

(1/3 the size of equivalent keyword set)

21

More experiments in paper

• Combining keyword and output bidding– E.g., a movie advertiser uses:

1. imdb.com wikipedia.org cover 80% of queries2. keyword1, keyword2, … cover 20% of queries

• Combined expressions– are as compact as output expressions– yield less (negative) spill than output expressions

Conclusions

• Output URL bidding can be implemented efficiently

• Advantages over keyword bidding– Bid compactness– More relevant queries (positive spill)

• Combining keyword and output bidding seems to be the most promising direction

Thank you!

Contact: papadimitriou@stanford.eduhttp://panagiotis.info